Add resources (machines) to the pool

From Supercomputación y Cálculo Científico UIS
Revision as of 19:22, 9 April 2015 by Ltorres (talk | contribs)


Logo_sc33.png

Add resources (machines) to the pool

In this section we describe all the administration tasks for the job scheduler OAR in the frontend node (Server) and in the compute nodes (Client)


Add the resources (new nodes) to OAR

1. Edit a file /tmp/nodos with the names of the new nodes (one per line)

File: /tmp/nodos
...
guane12
guane17
.
.
...

2. Then run the following command

oar_resources_init /tmp/nodos


3. Now run

source /tmp/oar_resources_init.cmd


4. Add GPU resources from nodes. Edit a file (GPUresources.cmd) with the following lines:

File: GPUresources.cmd
...
oarnodesetting --sql "core=169" -p gpu=YES -p gpunum=57 -p gputype=M2075
oarnodesetting --sql "core=170" -p gpu=YES -p gpunum=57 -p gputype=M2075
oarnodesetting --sql "core=171" -p gpu=YES -p gpunum=57 -p gputype=M2075
.
.
...
NOTE: Each new core must correspond with the new GPU. In guane, for instance, where each node has 24 cores and 8 GPU, each GPU has been assigned to 3 cores. The following script can be used to generate that file. Choose the counters accordingly to the situation: i, j, k, gpu.
File: script.sh
# Este script genera un (CPUresources.cmd) que contiene
# las líneas necesarias para agregar los recursos GPU de un
# nodo a la BD de OAR
# En el caso de guane, hay 24 cores por nodo y 8 GPUs por nodo
# Entonces se asigna un GPU cada 3 cores.
#!/bin/bash

i=169
j=170
k=171
gpu=57

while [ $gpu -lt 65 ]
   do
      echo "oarnodesetting --sql \"core=$i\" -p gpu=YES -p gpunum=$gpu -p gputype=M2075" >> GPUresources.cmd
      echo "oarnodesetting --sql \"core=$j\" -p gpu=YES -p gpunum=$gpu -p gputype=M2075" >> GPUresources.cmd
      echo "oarnodesetting --sql \"core=$k\" -p gpu=YES -p gpunum=$gpu -p gputype=M2075" >> GPUresources.cmd
      i=`echo $i + 3 | bc`
      j=`echo $j + 3 | bc`
      k=`echo $k + 3 | bc`
      gpu=`echo $gpu + 1 | bc`
   done

5. Add the resources with the following command:

source GPUresources.cmd